A Research On Algorithm For Side-Scan Sonar Image ROI Extraction

Extraction of regions of interest of side-scan sonar image is an vital part of its processing and vital to target recogniton. In this paper extraction of regions of interset of side-scan sonar image was studied from border line, regions and clustering.The method of extracting regions of interest from border line was traditional. Generally speaking, the border line information of side-scan sonar was not very strong, speciallly the shadow area of targets and the uniformity of high-light area was not very good. Most edge detection operator has a strong response not only on the edge of side-scan sonar, but also on high-light area. And the response was not strong on the edge of shadow area. All these reasons lead to the hard work for combining the edge point. So it is hard work to extract regions of interests from edge detection.Extractiong of regions of interests from region was an important research direction. Generally the whole information of regions of interests was considered, the result was credible. Firstly, a method based on iterative segmentation was studied. The property of segmentation result determined threshold for segmentation. Secondly, using the seed regions of shadow region and high-light area, adopting the similarity criterion, the regions of interest are extracted based on region growing. Thirdly, a Fourier analysis method for target detection in side-scan image. Using horizon line, vertical line, slope 45°line and slope 135°line spectral mask, the projection of side-scan sonar image in these orientation were obtained. Based on the sum of energy of four orientation projections, a threshold was set to judge target existing, and the position of target was also given.In the third part of this paper extraction of region of interest from clustering analysis was studied. Firstly, using the average of pixel as a feature, based on the known information, the pixvel was divided into three category, using the variance within certain groop as the objective function, the region of interest of side scan sonar was extracted for clustering analysis. Considering region connection, the result was perfected. Secondly, using average of pixvel, region roughness, region area and fractal dimension as features vectors, based on K-mean clusering, iterative clustering and fuzzy clustering, the region of interest was extracted.